DeepSeek, a Chinese startup has seemingly become the talk of the AI town, especially due to its R1 model which surpasses OpenAI’s o1 reasoning model capabilities across math, science, and coding at 3% of the cost.
Unlike other AI models from companies like Microsoft and OpenAI, which require exorbitant resources and vigorous training, the Chinese startup’s model is open-source. This allows interested parties to study and replicate its development at a fraction of the cost of sophisticated AI models. Benchmarks demonstrate impressive performance, outperforming sought-after models like GPT-4o, Gemini, and Claude.
As expected, the findings have caused a buzz in the AI community, with key players in the landscape astonished by its performance while still maintaining cost-effectiveness. Even Meta’s AI lead scientist, Yann LeCun has weighed in on DeepSeek’s AI milestone (via Business Insider):
“They came up with new ideas and built them on top of other people’s work. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source.”
DeepSeek has seemingly prompted doubts over US dominance in the AI landscape, causing stocks to plummet shortly after OpenAI and SoftBank announced their commitment to develop advanced AI infrastructure across the US with a $500 billion bet on the Stargate project.
According to LeCun, DeepSeek’s AI breakthrough goes beyond the long-standing rivalry and competition between China and the US. It reiterates the importance of open-source AI models, potentially bringing back OpenAI’s founding mission to life. The AI scientist further indicated that “open-source models surpass proprietary ones.”
DeepSeek’s R1 is arguably one of the few open-sourced models, alongside Meta’s Llama. Last year, OpenAI CEO Sam Altman indicated that the startup plans to release open-source products. Ironically, the AI firm was founded as an open-source company but has seemingly veered off this trail. The executive claimed it’s easier to meet safety thresholds by keeping advanced AI models closed-source.
As it happens, DeepSeek launched its app last week, quickly gaining popularity among users and dethroning OpenAI’s vast popular ChatGPT to become the most downloaded free app in the US. Elsewhere, Nvidia, Microsoft, and Meta have been hit by major share price drops following DeepSeek’s buzz in the AI community.
link